Tag
The article argues that retrieval-augmented generation (RAG) is often misapplied in AI systems, where the real issue is context curation rather than retrieval. It suggests that RAG is only truly beneficial for large, frequently changing corpora.
This article argues that PostgreSQL is a versatile database solution capable of replacing multiple specialized technologies such as search engines, message queues, and caches, thereby simplifying IT setups.
The article argues that vector databases alone act as data graveyards without true memory capabilities, and describes building a local AI memory layer with contradiction detection and forgetting cycles on top of Actian VectorAI DB.
Introduces the open-source project RAG from Scratch, which fully breaks down the RAG pipeline through step-by-step local code experiments, covering text chunking, vectorization, retrieval, reranking, query rewriting and other advanced strategies, helping developers understand RAG implementation from the ground up.
An educational article from MachineLearningMastery covering seven chunking strategies for RAG pipelines, from fixed-size token chunking to advanced semantic and hierarchical approaches, with guidance on when to use each.
Pathway's open-source llm-app is a framework for building enterprise-grade RAG systems. It supports real-time data sync, built-in vector retrieval, and comes with ready-made cloud templates. It has earned over 59,000 stars on GitHub.
This paper introduces ScalableRAG, a retrieval-augmented generation method that achieves high accuracy without any ingestion costs (no vector database or knowledge graph) by using regex-based set creation and aggregative reasoning. It outperforms baselines on multiple datasets and also presents a limited-ingestion variant for further accuracy improvements.
HARP is a training-free interpretability method that uses an LLM agent with a vector database of activations and tools for manipulation, outperforming training-based methods on concept discovery, detection, steering, and secret elicitation.
A tweet discusses how AI agents slow down over time because their vector databases are optimized for static data, but real agents continuously add new data, causing performance degradation.
A thread warns that choosing a vector database solely on speed benchmarks can be a trap for AI agents, which have constant write workloads unlike RAG's read-heavy patterns. It recommends specific databases based on use case, such as Qdrant for agent memory and pgvector for under 10M vectors on Postgres.
This tool indexes PDF and Markdown files locally for semantic search and AI enrichment, using Ollama for embeddings and LLM processing, and provides a CLI and MCP server for integration with AI assistants.
Memvid is an open-source tool that packages agent memory into mp4 files, offering significant performance improvements over traditional RAG pipelines and vector databases.
Alibaba has open-sourced the vector database zvec, which can be installed directly via pip – lightweight and easy to use.
Alibaba released Zvec, a fully open-source vector database (Apache 2.0) that can be installed via pip and supports dense, sparse, and hybrid search for RAG applications, processing billions of vectors in milliseconds.
The author criticizes the RAG and vector database approach, proposing that the correct methods include using memory correctly, chunking and indexing, summarizing, providing search tools for agents, and using SRAM-only inference services such as Groq and Cerebras.
A clear explanation of Standard RAG, Graph RAG, and Agentic RAG, covering their differences, use cases, and how they handle single-hop vs. multi-hop queries.
Mixedbread AI introduces asymmetric quantization for late interaction retrieval, achieving 32x storage reduction with minimal quality loss by storing document vectors as binary signs while keeping query vectors high-precision, making late interaction practical for billion-scale production systems.
A modular, high-performance Rust implementation of GraphRAG (Graph-based Retrieval Augmented Generation) with support for WebGPU acceleration and three deployment architectures: server-only, WASM-only (client-side), and hybrid.
The author built a fully offline AI agent using local embedding models, Llama via Ollama, and VectorAI DB to address the risks of cloud-dependent AI. The agent runs on an 8GB MacBook, processes sensitive documents, and maintains memory across sessions.
A detailed technical query about building a local document RAG system covering storage, ingestion, query, and highlighting, seeking advice on vector databases, GraphRAG feasibility, and document highlighting implementations.